Wild bees are widely distributed and effective pollinators, yet they face significant threats such as degradation of forests. Forest restoration has been advocated as a strategy to mitigate these threats and stabilize biodiversity. However, there is a lack of understanding of the ecological consequences of forest restoration on bee diversity, particularly regarding interactions with tree diversity and microenvironment. Using data from the world’s largest tree diversity experiment (BEF-China), this study examines how tree species richness, canopy cover, understory vegetation, and microclimatic conditions affect bee diversity in the context of forest restoration. Our analysis of bee diversity data (8,341 individuals from 79 species) revealed that these biotic factors had distinct effects on three dimensions of bee diversity. Specifically, canopy cover had a negative effect on bee taxonomic diversity but a positive effect on phylogenetic and functional diversity. However, these patterns were reversed when cover of understory vegetation was accounted for. Moreover, tree species richness exerted an indirect influence on bee diversity through understory microenvironment. Our findings provide nuance into how tree species richness shapes bee communities via vegetation cover and microclimate, which is informative on habitat characteristics in forest restoration and conservation that better enable the safeguarding of pollinators.

Study sites

The study was carried out in the BEF-China Biodiversity Experiment which is located in forests of Jiangxi province, a subtropical region (29°08′–29°11′N, 117°90′–117°93′E). For this study, 66 plots were selected randomly for the two sites (33 plots for each site), which covered a tree species richness gradient from 0 to 24 (for each site, 1 plot without artificial planted trees, 16 plots with monocultures, 8 plots with 2 species, 4 plots with 4 species, 2 plots with 8 species, 1 plot with 16 species and 1 plot with 24 species).

Environmental factors

Four quadrats were set on the diagonal (next to the traps) and one was set in the centre of each plot to measure the coverage of understory vegetation, of which the size was 1 m*1 m (see Fig S1). Above each quadrat, canopy cover was recorded through hemispherical pictures taken at 1 m above ground with a 180-mm fish eye lens and then calculated as percentage of black area of total image size using Gap Light Analyzer 2.0 (Frazer et al., 1999). Then, the mean of the canopy cover and understory coverage were calculated for each plot. For microclimate, temperature and humidity data were recorded in 15-minute intervals with sensors mounted 1 m above ground (Shandong Renke Control Technology Co., Ltd., China). The monthly temperature and humidity during the sampling events were calculated to represent understory microclimate.

Bee sampling and molecular work

Blue vane traps and three-colored pan traps were both used to collect bees in June and September of 2022, and April, June and September of 2023. Two vane traps were put at a distance of 1.5 m from ground, while two groups of pan traps (yellow, white and blue) were placed at a distance of 0.5 m on the second diagonal of each plot. Each trap was partially filled with water, with few drops of detergent to break the water surface tension. For each collection event, samples in the traps were collected after about 24 hours, conducted three times (totally 72 h). COI DNA barcode region (~650 bp) was amplified using universal primer pairs (LCO1490: GGTCA ACAAA TCATA AAGAT ATTGG and HCOout: CCAGG TAAAA TTAAA ATATA AACTT C). Molecular delimitation and taxonomic assignment followed the pipeline of Xie et at. (2023) (Xie et al., 2023).

Bee functional traits

Functional traits of our captured bees were determined either by new measurement (e.g. morphological traits) or phylogeny-based prediction using species-level reference data of phylogeny terminals and life history records. We measured morphometric traits of bee individuals using a Zeiss Discovery V20 stereomicroscope. If the length exceeded the microscope range, we used a digital calliper. Five such traits were used, body length (BL), head width (HW), hair length (HL), inter-tegular distance (ITD) and fore-wing length (fWL). Life history traits were also integrated via literature records, and then assigned to MOTU. These included lecty (floral range), sociality, parasitism and nest location (implicated in the utilization of resources). Missing traits were predicted and assigned by modeling trait evolution along branches of the phylogeny. A more comprehensive description of trait prediction for phylogenetically placed MOTU can be found in Xie et al. (2023).